A method for identifying illegal waste disposal based on image recognition algorithms

Abstract

This invention relates to the application of image recognition in waste disposal, specifically a method for identifying waste disposal violations based on image recognition algorithms. The method includes: capturing images of the area surrounding a waste bin in real time using a camera; detecting the similarity between adjacent frames using a structural similarity algorithm; detecting human and waste bag targets; analyzing the movement trajectory of the waste bag using a target tracking algorithm; determining whether the waste bag has been correctly disposed of in the waste bin; and triggering an alarm mechanism if the waste bag has not been correctly disposed of. This phased, systematic process reduces blind detection and unnecessary computation, improving overall efficiency. Thanks to the real-time capture by the camera, the system can obtain images immediately and quickly identify human bodies, waste bags, and disposal actions in subsequent stages. By separately detecting human and waste bag targets, accurate identification can be achieved in complex scenarios such as multiple people appearing simultaneously or multiple waste bags appearing simultaneously, providing richer basic data for motion trajectory analysis.

Description

Technical Field

[0001] This invention relates to the application of image recognition in waste disposal, specifically a method for identifying illegal waste disposal behavior based on image recognition algorithms. Background Technology

[0002] As waste sorting policies are gradually implemented and popularized across my country, people's awareness and enthusiasm for waste sorting are constantly increasing. Many communities and public places have begun to set up dedicated waste sorting and disposal points, equipped with corresponding guidance signs to help residents understand and correctly dispose of different types of waste. However, in actual implementation, some residents still violate regulations by mixing or improperly disposing of waste due to a lack of sufficient environmental awareness or unfamiliarity with waste sorting standards. These violations not only undermine the overall effectiveness of waste sorting but also create additional burdens for subsequent processing and recycling, affecting the efficiency of waste sorting and disposal.

[0003] Currently, traditional methods of supervising waste disposal often rely on on-site inspections by human managers or volunteers, or on video surveillance followed by manual review and judgment. This approach is not only costly and inefficient, but also prone to omissions and subjective human error. The pressure on manual monitoring is increasing daily, especially in large communities or areas with high population density. To improve the accuracy and real-time nature of monitoring, there is an urgent need for intelligent and automated methods to assist in supervision and management, reducing labor costs while significantly improving monitoring efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying illegal waste disposal based on image recognition algorithms, so as to solve the technical problems mentioned in the background.

[0005] Based on the above ideas, the present invention provides the following technical solution:

[0006] A method for identifying illegal waste disposal based on image recognition algorithms, comprising:

[0007] S1. Capture images of the area around the trash can in real time using a camera;

[0008] S2. Detect the similarity between adjacent frames based on structural similarity algorithm;

[0009] S3. Detect human and garbage bag targets;

[0010] S4. Analyze the movement trajectory of the garbage bag using a target tracking algorithm;

[0011] S5. Determine whether the garbage bag has been correctly disposed of in the garbage bin;

[0012] S6. If the garbage bag is not disposed of correctly, an alarm mechanism will be triggered.

[0013] Using camera-captured images as the entry point, the system performs hierarchical processing (S2-S5) on frame similarity, person-bag target detection, and bag trajectory analysis to ultimately determine whether the garbage was correctly disposed of and trigger an alarm. This phased, systematic process reduces blind detection and unnecessary computation, improving overall efficiency. Thanks to the real-time capture of the camera, the system can obtain images immediately and quickly identify people, garbage bags, and disposal actions in subsequent stages. Compared to purely manual inspections or simple video playback, this method significantly shortens reaction time and reduces the risk of human error. By detecting people and garbage bags separately, accurate identification can be achieved in complex scenarios such as multiple people or multiple garbage bags appearing simultaneously, providing richer basic data for subsequent motion trajectory analysis. It can be applied not only to urban community garbage bin areas but also to campuses, office buildings, public places, and can be integrated with other image processing modules or IoT platforms to form a broader smart management system.

[0014] Preferably, the threshold of the structural similarity algorithm in S2 is a dynamic parameter that is adjusted in real time according to scene characteristics, and is used to accurately detect significant changes between image frames.

[0015] Under varying lighting, weather conditions, and environmental noise, fixed thresholds often fail to guarantee optimal detection results. By using the SSIM (Structural Similarity) threshold as a dynamic parameter, adjustments can be made based on real-time scene brightness, background texture, and other factors, flexibly adapting to complex environments such as day / night cycles and varying weather conditions. As the threshold updates in real-time with the scene, it accurately identifies frames with significant differences, only proceeding to subsequent human or garbage bag detection processes when similarity drops significantly. This greatly reduces false triggers and computational overhead in inactive states. Using a dynamic threshold avoids premature or delayed activation of the recognition module in low-light or high-noise conditions, reducing both missed detections (actual movement being ignored) and false detections (noise fluctuations being mistaken for behavioral changes).

[0016] Preferably, when the structural similarity algorithm detects that the similarity value of adjacent frame images is lower than a preset threshold, the system enters the active state from the standby state and starts the image analysis module;

[0017] The structural similarity calculation formula is used to measure the similarity between two adjacent image frames. The structural similarity calculation formula is as follows:

[0018]

[0019] The formula measures the similarity of image structures. The closer the value is to 1, the more similar the two images are, and the closer it is to 0, the greater the difference.

[0020] in:

[0021] and Let x and y represent the mean values ​​of the image, respectively.

[0022] and Let x and y represent the variances of the image, respectively.

[0023] Represents the covariance of the x and y axes of the image;

[0024] and is the stability constant.

[0025] In standby mode, the system performs only low-cost SSIM calculations. Only when significant inter-frame differences are detected (low similarity values) will more in-depth analysis modules (human body recognition, garbage bag detection, etc.) be activated. This approach ensures real-time monitoring while effectively avoiding continuous high hardware load.

[0026] Subsequent steps are only triggered when the image undergoes significant changes at the structural and textural levels, thus reducing the frequency of interference for administrators.

[0027] Preferably, step S3 uses a human detection algorithm to determine whether a human target exists. If no human target is detected within a time threshold, the system returns to standby mode.

[0028] In active mode, human detection is performed for a limited time. If no human is detected, it indicates that the change in the image was not caused by the user, and the system will quickly return to standby mode to avoid continuous idle operation and wasting computing power.

[0029] Flickering lights, small animals in the environment, and leaves blowing in the wind can also cause SSIM differences. Subsequent processes are only triggered when a human body is detected, thus reducing the chance of misjudgment in "non-human-driven" scenarios.

[0030] At night or during periods of low human activity, if no human is detected for an extended period, the system will remain in low-power standby mode most of the time, which helps extend the lifespan of the device and reduce operating costs.

[0031] Preferably, after detecting a human target, the system enters an early warning state and identifies the garbage bag target through the garbage bag detection model. At the same time, it analyzes the distance and overlap between the human body and the garbage bag detection frame to determine whether to start the garbage disposal behavior.

[0032] Once a human body is detected, the system first issues an "early warning" rather than immediately determining a violation, allowing a buffer period for subsequent garbage bag inspection. This enables the system to manage processes more finely and improves the accuracy of recognizing both brief and continuous actions.

[0033] The garbage bag detection model is used to further identify whether there is a garbage bag in the picture, and the relative position between the two (distance, IoU overlap, etc.) is combined to determine whether it is in the key action stage such as "taking the bag" or "preparing to dispose of".

[0034] If the human body and the garbage bag do not present a reasonable spatial relationship (such as being far apart or not overlapping), it can be considered that no delivery behavior has been triggered, and the system will not consume additional computing power to enter trajectory analysis; the next step of tracking will only be started when a suspected delivery action occurs.

[0035] Preferably, the target tracking algorithm combines Kalman filtering and the Hungarian algorithm, and includes the following steps:

[0036] Kalman filtering predicts the next position of the target;

[0037] Minimize the matching cost between the target and the detection box using the Hungarian algorithm;

[0038] Update the forecast results and continue to track the target;

[0039] The Kalman filter's predicted state estimation formula includes:

[0040]

[0041] The predicted state estimation formula uses the system's dynamic model to predict the target state at the current moment, and the dynamic model includes position and velocity models.

[0042] The Kalman filter prediction covariance formula includes:

[0043]

[0044] The prediction covariance formula is based on the previous state estimate and the magnitude of the system noise prediction error.

[0045] in:

[0046] The predicted state vector;

[0047] The predicted covariance matrix;

[0048] Q is the process noise covariance matrix;

[0049] The Hungarian algorithm determines the optimal match by minimizing the cost matrix, including:

[0050] Define a cost matrix C, where each element c(ij) represents the matching cost between target i and detection box j.

[0051] By using the state prediction formula, the target position (including position, velocity and other components) can be estimated before the next frame arrives, thus solving the problem of lost detection boxes caused by short-term occlusion or jitter.

[0052] The uncertainty is quantitatively measured using the covariance prediction formula and then fused with the observations in subsequent update steps to obtain a more accurate target location.

[0053] This "prediction-update" cycle improves the robustness and effectiveness of tracking.

[0054] The optimal matching of the Hungarian algorithm: Construct a cost matrix C, whose elements can measure the similarity between historical target i and new detection box j based on Euclidean distance, IoU difference, etc.

[0055] By minimizing the cost, the "one-to-many" or "many-to-one" conflict in a multi-target environment is resolved, ensuring an accurate one-to-one correspondence between targets and detection boxes.

[0056] This mechanism can effectively distinguish different targets when multiple people or multiple garbage bags are present, making subsequent trajectory analysis and violation identification more reliable.

[0057] Multi-target tracking efficiency: Combining Kalman filtering and the Hungarian algorithm, it can process multi-person delivery scenarios in parallel, ensuring the tracking accuracy of each "person-bag" interaction in complex situations.

[0058] Preferably, when a garbage bag is not properly disposed of in the garbage bin and the distance between it and the detection frame of a human target exceeds a set threshold, the system triggers an alarm mechanism and records the image data at the time of the violation.

[0059] The fact that a garbage bag is not put into the bin can be determined by detecting the degree of overlap or coordinates between the final position of the garbage bag and the area of ​​the garbage bin.

[0060] The fact that the human detection box is far from the bag indicates that the person who placed the package has left, and that the person who placed the package has not taken any remedial action.

[0061] When both conditions are met simultaneously, it is possible to more accurately determine the occurrence of violations in ad placement.

[0062] If the person placing the item moves away slightly but then returns to place it again, no alarm will be triggered as long as the detected human body is still within the threshold distance. This effectively avoids false alarms caused by a momentary lapse followed by a correction.

[0063] Once the violation conditions are met, the system will immediately issue an alarm, which can not only quickly notify the management personnel to handle the situation, but also reserve an interface for subsequent automatic warnings or scene broadcasts, helping to reduce human negligence.

[0064] Preferably, the record of violations includes: screenshots of the human body and the garbage bag detection frame, timestamps of the violation and scene images, and the data is uploaded to the management platform for subsequent review and processing.

[0065] The system saves and uploads screenshots of the human body and garbage bag detection frames, timestamps, and the entire scene together, forming a complete chain of evidence, which facilitates subsequent inquiries and penalties by management departments or higher-level platforms.

[0066] Without the need for manual recording or exporting of videos, the system automatically packages key images and marks the event time when a violation is triggered, and sends the data to the management platform, greatly improving supervision efficiency.

[0067] The collected images and time information of violations can be used to statistically analyze peak periods of violations and common types of incorrect waste disposal, providing objective data support for subsequent optimization of waste disposal site design and public education.

[0068] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0069] By first detecting inter-frame differences using structural similarity (SSIM), and then combining this with human and garbage bag detection, Kalman filtering, and the Hungarian algorithm for multi-target tracking, the system can quickly determine whether the garbage bag is disposed of correctly and trigger an alarm. This phased and modular design not only ensures recognition accuracy but also significantly improves the system's real-time monitoring capabilities and response speed for violations.

[0070] In standby mode, the system performs only low-cost inter-frame similarity monitoring. More complex detection and tracking algorithms are only activated when significant changes are detected, greatly reducing unnecessary computation and hardware power consumption. Simultaneously, the dynamic threshold setting can be adjusted in real time according to different lighting conditions and environmental noise, thus maintaining a low false alarm rate and false negative rate in various scenarios.

[0071] Once a violation is detected and the person leaving the site has departed, the system automatically records a screenshot of the person and the garbage bag detection frame, along with a timestamp and a scene image, and uploads it to the management platform. This not only allows for timely reporting of abnormal situations to relevant departments but also generates traceable data, providing a reliable basis for subsequent review, penalties, and optimization of management strategies. Attached Figure Description

[0072] Figure 1 This is a system flowchart of a method for identifying illegal waste disposal based on image recognition algorithms according to the present invention.

[0073] Figure 2 This is a schematic diagram of the inter-frame change detection of the SSIM algorithm in the garbage disposal violation identification method based on image recognition algorithm of the present invention.

[0074] Figure 3 This is a schematic diagram illustrating the relationship between the human body and the garbage bag detection frame in a garbage disposal violation identification method based on image recognition algorithm according to the present invention.

[0075] Figure 4 This is a schematic diagram of the Kalman filter prediction and update process of a waste disposal violation identification method based on image recognition algorithm according to the present invention. Detailed Implementation

[0076] A method for identifying illegal waste disposal based on image recognition algorithms, comprising:

[0077] S1. Capture images of the area around the trash can in real time using a camera;

[0078] S2. Detect the similarity between adjacent frames based on structural similarity algorithm;

[0079] S3. Detect human and garbage bag targets;

[0080] S4. Analyze the movement trajectory of the garbage bag using a target tracking algorithm;

[0081] S5. Determine whether the garbage bag has been correctly disposed of in the garbage bin;

[0082] S6. If the garbage bag is not disposed of correctly, an alarm mechanism will be triggered.

[0083] Using camera-captured images as the entry point, the system performs hierarchical processing (S2-S5) on frame similarity, person-bag target detection, and bag trajectory analysis to ultimately determine whether the garbage was correctly disposed of and trigger an alarm. This phased, systematic process reduces blind detection and unnecessary computation, improving overall efficiency. Thanks to the real-time capture of the camera, the system can obtain images immediately and quickly identify people, garbage bags, and disposal actions in subsequent stages. Compared to purely manual inspections or simple video playback, this method significantly shortens reaction time and reduces the risk of human error. By detecting people and garbage bags separately, accurate identification can be achieved in complex scenarios such as multiple people or multiple garbage bags appearing simultaneously, providing richer basic data for subsequent motion trajectory analysis. It can be applied not only to urban community garbage bin areas but also to campuses, office buildings, public places, and can be integrated with other image processing modules or IoT platforms to form a broader smart management system.

[0084] Specifically, the threshold of the structural similarity algorithm in S2 is a dynamic parameter that is adjusted in real time according to scene characteristics, and is used to accurately detect significant changes between image frames.

[0085] Under varying lighting, weather conditions, and environmental noise, fixed thresholds often fail to guarantee optimal detection results. By using the SSIM (Structural Similarity) threshold as a dynamic parameter, adjustments can be made based on real-time scene brightness, background texture, and other factors, flexibly adapting to complex environments such as day / night cycles and varying weather conditions. As the threshold updates in real-time with the scene, it accurately identifies frames with significant differences, only proceeding to subsequent human or garbage bag detection processes when similarity drops significantly. This greatly reduces false triggers and computational overhead in inactive states. Using a dynamic threshold avoids premature or delayed activation of the recognition module in low-light or high-noise conditions, reducing both missed detections (actual movement being ignored) and false detections (noise fluctuations being mistaken for behavioral changes).

[0086] Specifically, when the structural similarity algorithm detects that the similarity value of adjacent frame images is lower than a preset threshold, the system enters the active state from the standby state and starts the image analysis module;

[0087] The structural similarity calculation formula is used to measure the similarity between two adjacent image frames. The structural similarity calculation formula is as follows:

[0088]

[0089] The formula measures the similarity of image structures. The closer the value is to 1, the more similar the two images are, and the closer it is to 0, the greater the difference.

[0090] in:

[0091] and Let x and y represent the mean values ​​of the image, respectively.

[0092] and Let x and y represent the variances of the image, respectively.

[0093] Represents the covariance of the x and y axes of the image;

[0094] and is the stability constant.

[0095] In standby mode, the system performs only low-cost SSIM calculations. Only when significant inter-frame differences are detected (low similarity values) will more in-depth analysis modules (human body recognition, garbage bag detection, etc.) be activated. This approach ensures real-time monitoring while effectively avoiding continuous high hardware load.

[0096] Subsequent steps are only triggered when the image undergoes significant changes at the structural and textural levels, thus reducing the frequency of interference for administrators.

[0097] Specifically, S3 uses a human detection algorithm to determine whether a human target exists. If no human target is detected within a time threshold, the system returns to standby mode.

[0098] In active mode, human detection is performed for a limited time. If no human is detected, it indicates that the change in the image was not caused by the user, and the system will quickly return to standby mode to avoid continuous idle operation and wasting computing power.

[0099] Flickering lights, small animals in the environment, and leaves blowing in the wind can also cause SSIM differences. Subsequent processes are only triggered when a human body is detected, thus reducing the chance of misjudgment in "non-human-driven" scenarios.

[0100] At night or during periods of low human activity, if no human is detected for an extended period, the system will remain in low-power standby mode most of the time, which helps extend the lifespan of the device and reduce operating costs.

[0101] Specifically, after detecting a human target, S3 enters an early warning state and identifies the garbage bag target through the garbage bag detection model. At the same time, it analyzes the distance and overlap between the human body and the garbage bag detection frame to determine whether to start garbage disposal behavior.

[0102] Once a human body is detected, the system first issues an "early warning" rather than immediately determining a violation, allowing a buffer period for subsequent garbage bag inspection. This enables the system to manage processes more finely and improves the accuracy of recognizing both brief and continuous actions.

[0103] The garbage bag detection model is used to further identify whether there is a garbage bag in the picture, and the relative position between the two (distance, IoU overlap, etc.) is combined to determine whether it is in the key action stage such as "taking the bag" or "preparing to dispose of".

[0104] If the human body and the garbage bag do not present a reasonable spatial relationship (such as being far apart or not overlapping), it can be considered that no delivery behavior has been triggered, and the system will not consume additional computing power to enter trajectory analysis; the next step of tracking will only be started when a suspected delivery action occurs.

[0105] Specifically, the target tracking algorithm combines Kalman filtering and the Hungarian algorithm, and includes the following steps:

[0106] Kalman filtering predicts the next position of the target;

[0107] Minimize the matching cost between the target and the detection box using the Hungarian algorithm;

[0108] Update the forecast results and continue to track the target;

[0109] The Kalman filter's predicted state estimation formula includes:

[0110]

[0111] The predicted state estimation formula uses the system's dynamic model to predict the target state at the current moment, and the dynamic model includes position and velocity models.

[0112] The Kalman filter prediction covariance formula includes:

[0113]

[0114] The prediction covariance formula is based on the previous state estimate and the magnitude of the system noise prediction error.

[0115] in:

[0116] The predicted state vector;

[0117] The predicted covariance matrix;

[0118] Q is the process noise covariance matrix;

[0119] The Hungarian algorithm determines the optimal match by minimizing the cost matrix, including:

[0120] Define a cost matrix C, where each element c(ij) represents the matching cost between target i and detection box j.

[0121] By using the state prediction formula, the target position (including position, velocity and other components) can be estimated before the next frame arrives, thus solving the problem of lost detection boxes caused by short-term occlusion or jitter.

[0122] The uncertainty is quantitatively measured using the covariance prediction formula and then fused with the observations in subsequent update steps to obtain a more accurate target location.

[0123] This "prediction-update" cycle improves the robustness and effectiveness of tracking.

[0124] The optimal matching of the Hungarian algorithm: Construct a cost matrix C, whose elements can measure the similarity between historical target i and new detection box j based on Euclidean distance, IoU difference, etc.

[0125] By minimizing the cost, the "one-to-many" or "many-to-one" conflict in a multi-target environment is resolved, ensuring an accurate one-to-one correspondence between targets and detection boxes.

[0126] This mechanism can effectively distinguish different targets when multiple people or multiple garbage bags are present, making subsequent trajectory analysis and violation identification more reliable.

[0127] Multi-target tracking efficiency: Combining Kalman filtering and the Hungarian algorithm, it can process multi-person delivery scenarios in parallel, ensuring the tracking accuracy of each "person-bag" interaction in complex situations.

[0128] Kalman gain calculation equation The Kalman gain is calculated based on the error between the prediction error and the actual measurement. This gain determines how the prediction results are adjusted to fit the measurement data.

[0129] State estimation update equation The predicted state is corrected by Kalman gain, and the target's state estimate is updated.

[0130] Covariance update equation Update the error covariance matrix to reflect the uncertainty of the current state estimate.

[0131] in:

[0132] K k Kalman gain;

[0133] H is the observation matrix;

[0134] R is the measurement noise covariance matrix;

[0135] Z k This is the observation vector.

[0136] Specifically, when a garbage bag is not properly disposed of in the garbage bin and the distance between it and the detection frame of a human target exceeds a set threshold, the system triggers an alarm mechanism and records the image data of the violation.

[0137] The fact that a garbage bag is not put into the bin can be determined by detecting the degree of overlap or coordinates between the final position of the garbage bag and the area of ​​the garbage bin.

[0138] The fact that the human detection box is far from the bag indicates that the person who placed the package has left, and that the person who placed the package has not taken any remedial action.

[0139] When both conditions are met simultaneously, it is possible to more accurately determine the occurrence of violations in ad placement.

[0140] If the person placing the item moves away slightly but then returns to place it again, no alarm will be triggered as long as the detected human body is still within the threshold distance. This effectively avoids false alarms caused by a momentary lapse followed by a correction.

[0141] Once the violation conditions are met, the system will immediately issue an alarm, which can not only quickly notify the management personnel to handle the situation, but also reserve an interface for subsequent automatic warnings or scene broadcasts, helping to reduce human negligence.

[0142] Specifically, the records of violations include: screenshots of the human body and the garbage bag detection frame, timestamps of the violations, and scene images. The data is uploaded to the management platform for subsequent review and processing.

[0143] The system saves and uploads screenshots of the human body and garbage bag detection frames, timestamps, and the entire scene together, forming a complete chain of evidence, which facilitates subsequent inquiries and penalties by management departments or higher-level platforms.

[0144] Without the need for manual recording or exporting of videos, the system automatically packages key images and marks the event time when a violation is triggered, and sends the data to the management platform, greatly improving supervision efficiency.

[0145] The collected images and time information of violations can be used to statistically analyze peak periods of violations and common types of incorrect waste disposal, providing objective data support for subsequent optimization of waste disposal site design and public education.

Claims

1. An image recognition algorithm-based garbage throwing violation behavior identification method, characterized in that, Comprise: S1, capture the image of the surrounding of the garbage can through the camera in real time; S2, detect the similarity of adjacent frame images based on the structural similarity algorithm; S3, detect the human body and garbage bag target; S4, analyze the motion trajectory of the garbage bag through the target tracking algorithm; S5, judge whether the garbage bag is correctly delivered into the garbage can; S6, if the garbage bag is not correctly delivered, trigger the alarm mechanism; The threshold value of the structural similarity algorithm in S2 is a dynamic parameter, which is adjusted in real time according to the scene characteristics, and is used to accurately detect the significant changes between image frames; When the structural similarity algorithm detects that the similarity value of adjacent frame images is lower than the preset threshold value, the system enters the active state from the standby state and starts the image analysis module; The structural similarity calculation formula is used to measure the similarity of adjacent two frame images, and the structural similarity calculation formula is: The formula measures the similarity of image structure, and the value closer to 1 indicates that the two images are more similar, and the closer to 0 indicates that the difference is greater; Wherein: and denote the mean of images x and y, respectively; and denote the variance of the images x and y, respectively; covariance of the images x and y; and Kstability is the stability constant; S3 judges whether there is a human target through a human body detection algorithm, if no human target is detected within a time threshold, the system returns to standby state; Wherein S3 detects the human target, the system enters the early warning state, and identifies the garbage bag target through the garbage bag detection model, analyzes the distance and overlap relationship between the human body and the garbage bag detection frame, and judges whether to start the garbage delivery behavior; The target tracking algorithm combines Kalman filtering and Hungarian algorithm, including the following steps: Kalman filtering predicts the next position of the target; The Hungarian algorithm minimizes the matching cost between the target and the detection frame; Update the prediction result and continue to track the target; The prediction state estimation formula of the Kalman filter predicts the target state at the current time using the dynamic model of the system, and the dynamic model includes position and speed model; The prediction covariance formula of the Kalman filter predicts the error size of the previous state estimation and the system noise, The Hungarian algorithm determines the optimal matching by minimizing the cost matrix, including: Define the cost matrix C, and the element c(ij) of the cost matrix C represents the matching cost of target i and detection frame j; When the garbage bag is not correctly delivered into the garbage can and the distance between the detection frame of the human target exceeds the set threshold value, the system triggers the alarm mechanism, and records the image data when the violation occurs. 2.The garbage throwing violation behavior recognition method based on image recognition algorithm according to claim 1, characterized in that, The record of the violation includes: the screenshot of the human body and the garbage bag detection frame, the timestamp of the violation and the scene image, and the data is uploaded to the management platform for subsequent review and processing.

Images